Mathematica Bohemica

Khaing Khaing Aye; Peng Yee Lee

The dual of the space of functions of bounded variation

Mathematica Bohemica, Vol. 131 (2006), No. 1, 1–9

Persistent URL: http://dml.cz/dmlcz/134078

Terms of use:
© Institute of Mathematics AS CR, 2006

Institute of Mathematics of the Czech Academy of Sciences provides access to digitized documents
strictly for personal use. Each copy of any part of this document must contain these Terms of use.

This document has been digitized, optimized for electronic delivery and
stamped with digital signature within the project DML-CZ: The Czech Digital
Mathematics Library http://dml.cz
131 (2006) MATHEMATICA BOHEMICA No. 1, 1–9

THE DUAL OF THE SPACE OF FUNCTIONS

OF BOUNDED VARIATION

Khaing Khaing Aye, Peng Yee Lee, Singapore

(Received May 31, 2005)

Dedicated to Prof. J. Kurzweil on the occasion of his 80th birthday

Abstract. In the paper, we show that the space of functions of bounded variation and
the space of regulated functions are, in some sense, the dual space of each other, involving
the Henstock-Kurzweil-Stieltjes integral.

Keywords: bounded variation, two-norm space, dual space, linear functional, Henstock
integral, Stieltjes integral, regulated function

MSC 2000 : 26A39, 26A42, 26A45, 46B26, 46E99

1. Introduction

The classical Riesz representation theorem is well known [5]. It states that a
continuous linear functional on the space C[a, b] of continuous functions on [a, b]
Rb
can be represented in terms of the Riemann-Stieltjes integral a f dg for f ∈ C[a, b]
where g is a function of bounded variation on [a, b]. In other words, the Riesz
representation theorem showed that the dual of C[a, b] is the space BV of functions
of bounded variation or more precisely a subspace of BV. However we fail to say in
whatever sense that the space C[a, b] is the dual of BV .
In 1966, Hildebrandt [9] has characterized continuous linear functionals on the
space BV regarding BV as a two-norm space. By a two-norm space we mean that
the space is endorsed with two-norm topology or two-norm convergence. More pre-
cisely, a sequence is convergent in the two-norm sense if the sequence is bounded in
one given norm and convergent in another given norm. Further Hildebrandt used the
left Cauchy integral [8, p. 87] among others. It is not explicit how his representation

1
theorem is connected with the Riesz theorem. In this paper, we reformulate Hilde-
brandt’s results in terms of the Henstock-Kurzweil-Stieltjes integral, and establish a
duality in which a part of it is the Riesz representation theorem. More precisely, let
BVN be the subspace of BV such that the condition f (x) = 12 [f (x+) + f (x−)] holds
for x ∈ (a, b) with f (a) = f (a+) and f (b) = f (b−) whenever f ∈ BVN . A regulated
function [2] on [a, b] is a function whose one-sided limits exist at every point of [a, b].
Let RF denote the space of regulated functions on [a, b] , and RFN the subspace of
RF satisfying the condition as in BVN . Regarding BVN as a two-norm space and
RFN a normed space, then the dual of BVN is RFN and vice versa. Note that C[a, b]
is a subspace of RFN and its dual is BVN . The fact that the dual of C [a, b] is BVN
can now embedded into the duality of RFN and BVN .
Let V (g; [a, b]) or V (g) denote the total variation of g on [a, b]. A function g is of
bounded variation on [a, b] or g ∈ BV if V (g) is finite. A sequence {gn } of functions
on [a, b] is said to be two-norm convergent to g in BV if V (gk ) 6 M for all k and gk
is uniformly convergent to g on [a, b]. The space BV is now a two-norm space with
the two-norm structure provided by the two-norm convergence in BV . Further, RF
is a normed space with norm given by kf k = sup {|f (x)| : x ∈ [a, b]}.

2. The Henstock-Stieltjes integral

We define the Henstock-Kurzweil-Stieltjes integral [6], [7] and prove a convergence

theorem. A real-valued function f is said to be Henstock-Kurzweil-Stieltjes integrable
to A with respect to g on [a, b] or (f, g) Henstock-Kurzweil-Stieltjes integrable on
[a, b] , if for every ε > 0, there exists a function δ(ξ) > 0 for ξ ∈ [a, b] such that for
any division D of [a, b] given by a = x0 < x1 < . . . < xn = b, with ξ1 , ξ2 , . . . , ξn
satisfying ξi ∈ [xi−1 , xi ] ⊂ (ξi − δ (ξi ) , ξi + δ (ξi )) for i = 1, 2, . . . , n, we have

X 
 n 
 f (ξ )[g(x ) − g(x )] − A  < ε.
 i i i−1 
i=1

For brevity, we write D = {(ξ, [u, v])} in which (ξ, [u, v]) denotes a typical point-
interval pair (ξi , [xi−1 , xi ]) in D, and also we write the Riemann sum in the form
P
(D) f (ξ)[g (v) − g (u)]. Here D is said to be δ-fine if the above condition holds [6],
[7]. That is, (f, g) is Henstock-Kurzweil-Stieltjes integrable to A on [a, b] if for every
ε > 0, there is a positive function δ such that for any δ-fine division D = {(ξ, [u, v])}
of [a, b] , we have

(∗) |s(f, g, D) − A| < ε

2
 P Rb
where s (f, g, D) = (D) f (ξ) [g (v) − g (u)]. For simplicity, we write A = a f dg.
A division D1 is said to be finer than a division D2 if every interval of D1 is a
subinterval of some interval of D2 . If g (x) = x, then the integral reduces to the
Henstock-Kurzweil integral. If δ is a constant, then it reduces to the Riemann-
Stieltjes integral. Moreover, if g(x) = x and δ is a constant, then it reduces to the
Riemann integral [10]. We remark that the Henstock-Kurzweil-Stieltjes integral of
(f, g) does allow both functions f and g to be discontinuous on the same side at a
point.
Let f be a real-valued function defined on [a, b] . A family F of Henstock-Kurzweil-
Stieltjes integrable functions (f, gk ) , k = 1, 2, . . ., is said to be equi-integrable on [a, b],
if the above inequality (∗) holds with g and A replaced by gk and the integral Ak of
gk , and δ being independent of k [7, p. 104]. It is clear that if f ∈ RF and g ∈ BV,
Rb
then a f dg exists, for reference see [3].

Lemma 1. Suppose f ∈ RF . If gk ∈ BV and V (gk ) 6 M for every k, then

{(f, gk ) : k = 1, 2, . . .} is equi-integrable on [a, b].

 . Since f ∈ RF, it is known [1] that for every ε > 0, there exists
a division D0 : a = x0 < x1 < . . . < xn = b such that |f (ξ) − f (η)| < ε for
ξ, η ∈ (xi−1 , xi ) , i = 1, 2, . . . , n. Fix i and put gk∗ (x) = gk (x) for every x ∈ (xi−1 , xi )
and gk∗ (xi−1 ) = gk (xi−1 +) , gk∗ (xi ) = gk (xi −). Then we have
Z xi Z xi
f dgk = f dgk∗ + f (xi−1 ) [gk (xi−1 +) − gk (xi−1 )]
xi−1 xi−1
+ f (xi ) [gk (xi ) − gk (xi −)].

Define a function δ (ξ) > 0 for every ξ ∈ [a, b] such that every δ-fine division D =
{(ξ, [u, v])} of [a, b] is finer than D0 . Take any δ-fine division D. Then we can write
D = D1 ∪ D2 ∪ . . . ∪ Dn where Di is a δ-fine division of [xi−1 , xi ]. Thus we have
 Z xi   Z xi 
   
s (f, gk , Di ) −  
f dgk  = s (f, gk , Di ) −
∗
f dgk∗  6 εV (gk ; [xi−1 , xi ]).

xi−1 xi−1

Hence for every δ-fine division D = {(ξ, [u, v])} of [a, b] ,

 Z b  X n  Z xi 
   
s (f, gk , D) − f dgk  6 s (f, g, Di ) − f dgk  6 εV (gk ; [a, b]) 6 εM
 
a i=1 xi−1

for every k. That is, {(f, gk ) : k = 1, 2, . . .} is equi-integrable on [a, b]. 

3
 Theorem 1. Suppose gn ∈ BV for n = 1, 2, . . . , and f ∈ RF . If gn is two-norm
Rb Rb Rb
convergent to g in BV, then a f dg exists and lim a f dgn = a f dg.
n→∞

 . Rb
It is clear that g ∈ BV and a f dg exists. Further, by Lemma 1,
{(f, gn ) : n = 1, 2, . . .} is equi-integrable on [a, b]. Then for every ε > 0, there exists
a function δ (ξ) for ξ ∈ [a, b] such that for every δ-fine division D = {(ξ, [u, v])} of [a, b]
Rb Rb
and for every n, we have |s (f, g, D) − a f dg| < ε and |s (f, gn , D) − a f dgn | < ε.
Since gn is uniformly convergent to g on [a, b] , we can prove that there exists D0
such that |s (f, gn , D0 ) − s (f, g, D0 )| < ε for large n. Therefore we have
Z b Z b  Z b 
   
 f dgn − f dg  6 f dgn − s (f, gn , D0 )  + |s (f, gn , D0 ) − s (f, g, D0 )|

a a a
Z b 
 

+ f dg − s (f, g, D0 )  < 3ε
a

for large n. 

3. Two-norm continuous linear functionals

A functional F defined on BV or RF is linear if F (αf + βg) = αF (f ) + βF (g)

for f, g ∈ BV or RF, and real α, β. A functional F defined on BV is said to be two-
norm continuous if F (gn ) → F (g) as n → ∞ whenever gn is two-norm convergent
Rb
to g in BV . Theorem 1 shows that if f ∈ RF, then F (g) = a f dg is two-norm
continuous in BV . A functional F defined on RF is continuous if F (fn ) → F (f ) as
n → ∞ whenever kfn − f k → 0 as n → ∞. In this section, we characterize the dual
or the two-norm dual of the space BV, that is, the space of all two-norm continuous
linear functionals on BV, in terms of the Henstock-Kurzweil-Stieltjes integral. We
define characteristic functions γt and δt for a fixed t ∈ [a, b]. When t ∈ (a, b) , define
γt (x) = 0 for every x ∈ [a, t), 21 when x = t and 1 for every x ∈ (t, b]. When t = a,
define γa (x) = 0 when x = a and 1 for every x ∈ (a, b]. When t = b, define γb (x) = 0
for every x ∈ [a, b) and 1 when x = b. We also define δt (x) = 1 when x = t and
0 otherwise. We now give a series of lemmas leading to the main theorem of this
paper.

Lemma 2. If F is a two-norm continuous linear functional defined on BV, then

f ∈ RFN where f (t) = F (γt ) for every t ∈ [a, b].

 .
Suppose f ∈ / RF . Then there exists a point x such that f (x+) or f (x−)
does not exist. Suppose f (x+) does not exist. Then there exist η > 0 and a sequence
of pairwise disjoint intervals [x0j , x00j ] on the same right side of x and approaching

4
 P
n
x such that |f (x00j ) − f (x0j )| > η for every j. Define gn = 1
n εj (γx00j − γx0j ) where
j=1
εj = sgn(f (x00j ) − f (x0j )). It is clearly seen that gn is uniformly convergent to zero on
[a, b] and V (gn ) 6 2 but
n n
1X 1X
F (gn ) = εj (F (γx00j ) − F (γx0j )) = |f (x00j ) − f (x0j )| > η,
n j=1 n j=1

for every n. That is, F is not a two-norm continuous linear functional on BV . There-
fore f (x+) and f (x−) must exist for every x. Hence f is a regulated function. In
view of the definition of γt , indeed f ∈ RFN . 

Lemma 3. Suppose F is a two-norm continuous linear functional defined on BV.

Rb
If f (t) = F (γt ) for every t ∈ [a, b] , then f ∈ RFN, the integral a f dg exists and
Rb
F (g) = a f dg for g ∈ BVN .

 Rb
. By Lemma 2, f ∈ RFN and consequently a f dg exists. In view of
Theorem 1, it is sufficient to prove Lemma 3 for the case when g is a step function
in BVN . Let ti , i = 1, . . . , n − 1, be the discontinuity points of g. Consider the case
when a = t0 < t1 < . . . < tn = b. Then
Z b n−1
X  n−1
X 
f dg = f (ti ) [g (ti +) − g (ti −)] = F γti · [g (ti +) − g (ti −)] .
a i=1 i=1

Here we note that

n−1
X
γti · [g (ti +) − g (ti −)] = g
i=1
Rb
and hence a
f dg = F (g) for the step function g ∈ BVN and consequently for any
g ∈ BVN . 
The linear space c0 [a, b] [5] is the space of all functions f defined on [a, b] such
that the set {t ∈ [a, b] : |f (t)| > ε} is finite for every ε > 0.

Lemma 4. If F is a two-norm continuous linear functional defined on BV, then

f ∈ c0 [a, b] where f (t) = F (δt ) for every t ∈ [a, b].

 .Suppose f ∈ / c0 [a, b]. Then there exists ε > 0 such that |f (ti )| > ε for
P
n
a sequence {ti }i>1 ⊂ [a, b]. Define gn = n1 εi δti , where εi = sgn f (ti ). Then we
i=1
obtain V (gn ) 6 2 and gn is uniformly convergent to 0 on [a, b] but
n n
1X 1X
F (gn ) = εi F (δti ) = |f (ti )| > ε for every n.
n i=1 n i=1

5
That is, F is not a two-norm continuous linear functional on BV . Therefore f ∈
c0 [a, b]. 

We define the normalized function g ∗ of g on [a, b] to be g ∗ (x) = 12 [g(x+) + g(x−)]

for x ∈ (a, b) , and g ∗ (a) = g (a+) , g ∗ (b) = g (b−). If g ∈ BV, then g ∗ ∈ BVN .
The main result of this paper is the following theorem.

Theorem 2. A functional F defined on BV is linear and two-norm continuous if

and only if there exist functions f1 ∈ RFN and f2 ∈ c0 [a, b] such that
Z b ∞
X
F (g) = f1 dg ∗ + [g (ti ) − g ∗ (ti )] f2 (ti )
a i=1

for every g ∈ BV, where ti , i = 1, 2, . . . , are the discontinuity points of g, and g ∗ the
normalized function of g.

 .
⇒: Put f1 (t) = F (γt ) for every t ∈ [a, b]. By Lemma 2, f1 ∈ RFN . Let
g ∈ BV and g ∗ the normalized function of g. Then

F (g) = F (g ∗ ) + F (g − g ∗ ).

Next, put f2 (t) = F (δt ) for every t ∈ [a, b]. By Lemma 4, f2 ∈ c0 [a, b]. Let ti ,
i = 1, 2, . . . , be the discontinuity points of g. Then for any n
X
∞ 
F (g − g ) = F
∗
[g (ti ) − g (ti )] δti
∗

i=1
n
X  X
∞ 
= [g (ti ) − g ∗ (ti )] f2 (ti ) + F [g (ti ) − g ∗ (ti )] δti .
i=1 i=n+1

P
∞
Since {f2 (ti )} is a null sequence, that is, a sequence converging to 0, and [g(ti )−
i=1
P
∞
g ∗ (ti )] absolutely convergent, the series [g (ti ) − g ∗ (ti )] f2 (ti ) converges.
i=1
P
∞
Further, [g (ti ) − g ∗ (ti )] δti is two-norm convergent to zero in BV as n → ∞.
i=n+1
Then, by the two-norm continuity of F, we have
 X
∞ 
lim F [g (ti ) − g ∗ (ti )]δti = 0.
n→∞
i=n+1

Together with Lemma 3, we obtain the representation of F.

6
 ⇐: Let
Z b
F1 (g) = f1 dg ∗ .
a

Take gn ∈ BV such that gn is two-norm convergent to g in BV . So does gn∗ . Since f1 ∈

RFN, by applying Lemma 1 and Theorem 1, {(f, gk ) : k = 1, 2, . . .} is equi-integrable
Rb Rb Rb
on [a, b] , a f1 dg ∗ exists and lim a f1 dgn∗ = a f1 dg ∗ . Hence lim F1 (gn ) = F1 (g).
n→∞ n→∞
That is, F1 (g) defines a two-norm continuous linear functional on BV .
Let
∞
X
F2 (g) = [g (ti ) − g ∗ (ti )]f2 (ti ).
i=1

Since f2 ∈ c0 [a, b] , for every ε > 0, the set X = {x ∈ [a, b] : |f2 (x)| > ε} is finite.
Consequently, we have |f2 (x)| < ε for every x ∈ / X. Since gn is two-norm convergent
to g in BV, for every ε > 0, there exists N such that for every n > N and for every
ξ ∈ [a, b] , we have
ε
|gn (ξ) − g(ξ)| <
n (X)

where n (X) denotes the number of the points in X. We may assume that |f2 (x)| 6
M1 for every x ∈ [a, b] and also we have V (gn ) 6 M2 for every n and V (g) 6 M2 .
Therefore for every n > N, we have
X X ε
(|gn (t) − g(t)| + |gn∗ (t) − g ∗ (t)|)|f2 (t)| 6 2 M1
n(X)
t∈X t∈X

and
X
(|gn (t) − gn∗ (t)| + |g(t) − g ∗ (t)|)|f2 (t)| 6 ε[V (gn ; [a, b]) + V (g; [a, b])].
t∈X
/

Therefore
 
 X 
 ([gn (t) − gn (t)] − [g(t) − g (t)])f2 (t) 6 2εM1 + ε(2M2 )
∗ ∗

t∈(a,b)

and hence lim F2 (gn ) = F2 (g). That is, F2 (g) defines a two-norm continuous linear
n→∞
functional on BV . Put F (g) = F1 (g) + F2 (g). Hence the proof is complete. 

As a special case, we obtain the following corollary.

7
 Corollary 1. A functional F defined on BVN is linear and two-norm continuous
if and only if there exists a function f ∈ RFN such that

Z b
F (g) = f dg for g ∈ BVN .
a

That is, the dual of the space BVN is the space RFN . Conversely, the dual of RFN
is BVN . This has been proved in [4] and is stated below. The proof is similar to that
of the Riesz theorem. A functional F on RFN is continuous if it is continuous with
respect to the norm in RFN .

Theorem 3 [4]. A functional F defined on RFN is linear and continuous if and

only if there exist a function g ∈ BV and a real number d such that

Z b
F (f ) = df (a) + g df for f ∈ RFN .
a

Therefore the spaces BVN and RFN are the duals of each other. A representation
theorem for continuous linear functionals on RF of the form like Theorem 2 is also
available [3], [12]. We shall not elaborate here.
Finally, the following Riesz representation theorem is a consequence of Theorem 3,
integration by parts, and the Hahn-Banach theorem, since C [a, b] is a closed subspace
of RFN .

Corollary 2 (Riesz representation theorem). A functional F defined on C [a, b]

is linear and continuous if and only if there exists a function g ∈ BVN such that

Z b
F (f ) = f dg for f ∈ C [a, b] .
a

Representation theorem for continuous linear functionals on RF was first given by

Schwabik [13]. For another relevant reference, see [14].

8
 References
[1] D. Franková: Regulated functions. Math. Bohem. 116 (1991), 20–59. Zbl 0724.26009
[2] J. Dieudonné: Foundations of Modern Analysis. New-York, 1960. Zbl 0176.00502
[3] K. K. Aye: The duals of some Banach spaces. Ph.D Thesis, Nanyang Technological Uni-
versity, 2002.
[4] M. Tvrdý: Linear bounded functionals on the space of regular regulated functions. Tatra
Mt. Math. Publ. 8 (1996), 203–210. Zbl 0920.46031
[5] P. Habala, P. Hájek, V. Zizler: Introduction to Banach Spaces. 1996.
[6] P. Y. Lee: Lanzhou Lectures on Henstock Integration. World Scientific, 1989.
Zbl 0699.26004
[7] P. Y. Lee, R. Výborný: The Integral: An Easy Approach after Kurzweil and Henstock.
Cambridge University Press, 2000. Zbl 0941.26003
[8] T. H. Hildebrandt: Introduction of the Theory of Integration. Academic Press, 1963.
Zbl 0112.28302
[9] T. H. Hildebrandt: Linear continuous functionals on the space (BV) with weak topologies.
Proc. Amer. Math. Soc. 17 (1966), 658–664. Zbl 0152.13604
[10] Tom M. Apostol: Mathematical Analysis. 1957. Zbl 0077.05501
[11] W. Orlicz: Linear Functional Analysis. World Scientific, 1992. Zbl 0799.46002
[12] K. K. Aye, P. Y. Lee: Orthogonally additive functionals on BV. Math. Bohem. 129
(2004), 411–419.
[13] Š. Schwabik: A survey of some new results for regulated functions. Seminario Brasileiro
de analise 28 (1988).
[14] M. Brokate, P. Krejčí: Duality in the space of regulated functions and the play operator.
Math. Z. 245 (2003), 667–668. Zbl 1055.46023

Authors’ addresses: Khaing Khaing Aye, (1) Mathematics and Mathematics Educa-
tion, National Institute of Education, Nanyang Technological University, 1, Nanyang Walk,
Singapore 637616, (2) Department of Engineering Mathematics, Yangon Technological Uni-
versity, Yangon, Myanmar, e-mail: wk.ygn@mptmail.net.mm; Peng Yee Lee, Mathematics
and Mathematics Education, National Institute of Education, Nanyang Technological Uni-
versity, 1, Nanyang Walk, Singapore 637616, e-mail: pylee@nie.edu.sg.